In a new post on the Pinterest Engineering blog, Pinterest’s team outlined the importance of variable signals within its Pin recommendations. It also underscored the impact that this had compared to only showing Pins that aligned directly with each users’ interest.
This audio is auto-generated. Please let us know if you have feedback.
In a new post on the Pinterest Engineering blog, Pinterest’s team outlined the importance of variable signals within its Pin recommendations. It also underscored the impact that this had compared to only showing Pins that aligned directly with each users’ interest.
As explained by Pinterest: “While earlier stages [of the feed recommendation process] mostly optimize for certain positive actions (e.g., saves) given an impression, the multi-objective optimization layer tackles a different problem: determining the best composition of a feed served to the user. This is critical as users tend to have lower intent when visiting Home Feed and their browsing behavior will be significantly impacted by what they see. For example, visually repetitive content is less engaging and is likely to reduce the user’s session length and the likelihood that a user will revisit Pinterest.”
Pinterest said that while showing content similar to what they’ve already searched for can have short-term engagement benefits, those signals quickly turn negative. In addition, those signals dilute user intent indicators, leading to an even less engaging feed experience over time.
Which makes sense. While users might be interested in a certain topic or product at one stage, that interest will shift over time. As such, using direct engagement as the core signal, without a more diverse display of recommendations, will mean less relevance.
The challenge then is how Pinterest can extrapolate behavioral indicators into other topics and ideas.
In order to address concerns with repetition, even when it is based on behavioral cues, Pinterest expanded this specific element within its recommendation process to ensure more diversity in content display.
Pinterest also added a range of additional signals to drive its recommendations, including expanding its interest signals beyond product type labels, in order to include more visual and text embeddings and “co-engagement patterns and neighborhood similarity.”
This expanded recommendation approach has led to improvements in user engagement by widening the scale of signals the app uses for recommendations to a much broader set of engagement indicators, including actions undertaken by other users.
It’s another element in Pinterest’s gradual evolution, which has seen it become a much more significant platform for product discovery. The platform currently has 619 million monthly active users, and those numbers are rising.
This insight also shows the importance of feed diversification and branching beyond what users engage with to ensure greater exposure, which leads to more interest.
Similar approaches could help to combat reinforcement bubbles in social media display.
